Storyline
AI Memory Wall and Rising Hardware Costs
Chipmakers and memory manufacturers at Hot Chips 2026 detail competing approaches to solving AI's memory bottleneck as Nvidia warns customers of double-digit price increases on AI servers due to memory supply constraints.
- Hot Chips 2026: Power and Memory, Not Raw Compute, Take Center Stage in AI Chip Designs
At Hot Chips 2026, Nvidia, Fujitsu, Arm, d-Matrix, Samsung and others presented new approaches to the two constraints increasingly limiting AI data centers: electrical power and memory bandwidth/capacity, with several companies proposing new memory tiers and dynamic power schemes rather than simply adding more compute.
- Chipmakers Race to Solve AI's Growing Memory Bottleneck as Costs Surge for Buyers
At Hot Chips 2026 and in recent industry disclosures, memory and chip companies described an intensifying 'memory wall' problem in AI computing, with rival approaches from d-Matrix, SK Hynix, Samsung, Micron, and Rapidus aiming to close the gap—while Nvidia has reportedly told customers to expect double-digit price increases on AI servers due to memory supply constraints.